In this research, we analyzed the relationship between semantics of utterances and movements of head in a natural dialogue and a task oriented one in Japanese. We are going to show that visual information such as head movement will be useful for managing a dialogue and reducing the vagueness of semantics. First we extracted the head movements calculated automatically in a natural conversation and going to indicate the role of it. After this we will show an analysis of head movements during a cooperative problem solving task to construct a natural dialogue system in which initiative of the conversation moves. We will show the effectiveness of using visual information in a multimodal dialogue system.